Market Excess Demand Functions
Author(s): Hugo Sonnenschein
Source: Econometrica, Vol. 40, No. 3 (May, 1972), pp. 549-563
Published by: The Econometric Society
Stable URL: http://www.jstor.org/stable/1913184
Accessed: 03-07-2016 17:28 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact [email protected].
Wiley, The Econometric Society are collaborating with JSTOR to digitize, preserve and extend access to
Econometrica
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
Econometrica, Vol. 40, No. 3 (May, 1972)
MARKET EXCESS DEMAND FUNCTIONS
BY HuGo SONNENSCHEIN1
The purpose of this paper is to investigate the structure of the class of market excess
demand functions which can be generated by aggregating individual utility maximizing
behavior. Among the results are: (i) in a region of the relative price domain an arbitrary
polynomial function can be generated as an excess demand function for a particular
commodity, and (ii) for any p in the relative price domain, a given configuration of excess
demands and rates of change in excess demand can be generated at p if and only if it is
consistent with Walras' Law.
1. INTRODUCTION
THE CONCEPT of a market excess demand function occupies a central role in the
explanation of value furnished by all models of the competitive mechanism. It is
elementary that these functions must be homogeneous of degree zero in all prices
and satisfy Walras' Law. Under a standard set of assumptions they will be continuous (see, for example, [2]). Walras' Law together with continuity guarantee
that excess demand functions must have a zero, or in more familiar terminology,
that equilibrium prices must exist (see, for example, [2]). Clearly this zero may be
unique (consider the case of a single individual with smooth indifference surfaces).
Beyond these facts, and a small number of results from the "stability" literature
(see [4] and [9]), very little is known about the structure of excess demand
functions. A sampling of the results presented here and their relationship to the
existing literature follows. We assume in this discussion that there are n commodities and that the price of the nth commodity is fixed at unity. Price vectors
are points in the positive orthant of an n - 1 dimensional Euclidean space.
(1) Can an arbitrary continuous function, defined on a compact subset C of the
interior of a positive orthant, be an excess demand function for some commodity
in a general equilibrium economy? To my knowledge this question has received
little attention, yet specific functional forms for aggregate demand relationships
are the starting point for a large body of empirical work. Despite the generally
accepted importance of basing statistical studies on consistent theoretical grounds,
it is interesting to note that there is no literature exploring whether, for example,
a linear aggregate excess demand function is theoretically possible.
'The research on which this paper is based was done at the Pennsylvania State University. I
appreciate the comments of the people who discussed the results with me at a large number of seminars.
However, I am especially indebted to two colleagues: C. Moler, University of Michigan, who got me
started on the proof of the lemma in the Appendix; and an anonymous referee, who in addition to
suggesting the title, streamlined two theorems and convinced me to dismiss a sideshow.
549
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
550
HUGO
SONNENSCHEIN
In Theorem 2 we prove that every polynomial on C is an excess demand function
for a specified commodity in some n commodity economy. As a consequence of a
classic mathematical result on approximation, this theorem has as a corollary the
fact that every continuous real-valued function is approximately an excess demand
function.
(2) Conditions are known under which competitive equilibrium is unique (see,
for example, [2]); however, there are virtually no other results which yield information on what the sets of equilibrium prices can look like.2 In Theorem 4 we establish
that given any compact subset C of the interior of a positive orthant, and almost
any finite set P of prices in C, there exists an economy whose set of competitive
equilibrium prices in C is P. To the extent that an arbitrary bounded set may be
approximated by a finite set of points, this may be stated by saying that every
bounded set is approximately a set of equilibrium prices for some economy.
(3) Samuelson in his Foundations [6] characterized the comparative statics
properties of the class of demand functions which arise from individual utility
maximizing behavior.3 If a demand function is derived from utility maximizing
behavior, then its matrix of substitution terms must be symmetric and negative
definite at each point in the domain of the demand function. Furthermore, if a
function of prices and income yields a substitution matrix which is symmetric and
negative definite, then it is possible to derive that function as a demand function;
that is, the function can be generated from an individual's utility maximizing
behavior. To interpret this theorem in a slightly different way: the only relationship
among prices, quantities demanded, and rates of change in quantities demanded,
that is implied by the utility hypothesis, is that the substitution matrix be symmetric
and negative definite when evaluated at each point in the domain of the demand
function.
In Theorem 5 we prove a local analogue of the above characterization for the
case of aggregate excess demand functions. It states that Walras' Law characterizes
the comparative statics properties of the class of excess demand functions which
arise from aggregating individual utility maximizing behavior. Informally, we
summarize this proposition by saying that Walras' Law is the only local comparative statics theorem of competitive equilibrium analysis. The proof is established
by showing that given any price vector p, any n - 1 numbers N(1), N(2),....
N(n - 1), and any (n - 1)2 numbers a(i,j), there exists a collection of consumers
(each consumer is a utility function and an initial holding) such that:
(4a) the excess demand for the ith commodity at prices p is N(i), i = 1, 2, ... .
n - 1 (the excess demand for the nth commodity is determined by Walras' Law),
and
(4b) the partial derivative of the ith excess demand function with respect to the
jth price, evaluated at p, is a(i, j), i, j = 1, 2, ... , n - 1 (the partial derivatives of
the nth excess demand function are determined by Walras' Law).
2 A paper by G. Debreu [1] is an exception.
3 See also [7]. This result has been generalized by Hurwicz and Uzawa [3], and alternative characterizations are established in the revealed preference literature.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
DEMAND
FUNCTIONS
551
In order to illustrate the power of this result, we apply it to demonstrate the
existence of a competitive equilibrium which is Hicksian stable but not dynamically
stable, and this solves a problem which has been outstanding for over twenty-five
years (see [5 and 8]).
2. PRELIMINARY NOTIONS
Let a denote the unit (n - 1) simplex; i.e., a = {(x1, X2, . . *x ):Xi > 0 for all
i and z xi = 1, where the sum is taken over all i, 1 < i < n}. X denotes the interior
of JIand, for 0<6< 1, A(b) = {(x1,x2,...,xn)eA :x i>6, i = 1,2,...n}.The
set of (n - 1)-tuples of positive real numbers is denoted by H, and H(b), 0 < 6 < 1,
is that subset of H whose points have all of their co-ordinates between 6 and 1/6.
The set of vectors in Rn (Euclidean n-dimensional space) whose co-ordinates are
all non-negative is denoted by Q and is called the commodity space. A point in the
commodity space is called a bundle. A utility function U is a continuous realvalued function defined on the commodity space satisfying these conditions:
(5) for all bundles x and y, if each co-ordinate of x is larger than the corresponding
co-ordinate of y, then U(x) > U(y), and
(6) for all bundles x, {x': U(x') > U(x)} is convex.
The ith consumer is represented as a utility function Ui and a bundle
C0i = (wCj, wC, ... ., w'). The bundle coi is called the bundle of initial holdings of
individual i. Demand functions are defined in the usual way on both a and H,
and the demand function of the ith individual is denoted by hi. His excess demand
function Ei is defined by Ei = h- - oi.
We will consider the case where Ei(p) is a singleton for each p E a (respectively
p E H). It is easily established that p. Ei(p) = 0 (respectively (p, 1). Ei(p) = 0) for all
p E 3 (respectively p E H). If (Ui, Wi) is a consumer for each i E J, let EJ = , Et,
where the sum is taken over i E J. (EJ, EJ, . . . , EJ) = EJ is called the aggregate
excess demand function. It maps a (respectively H) into Rn. One can prove that
EJ satisfies Walras' Law; i.e., for each p E a (respectively p E H) we have p * EJ(p) = 0
(respectively (p, 1) . EJ(p) = 0).
3. A READER'S GUIDE
The analysis proceeds as follows. Let 0 < 6 < 1. In Theorem 1 we prove that
any real-valued function of one variable, which has a continuous derivative for
prices between 6 and 1/6, is an excess demand function on [6, 1/6] for the first
commodity in some two person two commodity economy (the excess demand for
the second commodity is determined by Walras' Law). By defining an appropriate
composite good, this fact is used to establish Lemma 1: Any function of n - 1
prices, which can be represented as a continuously differentiable function of a
linear function of the n - 1 prices on H(b), is the excess demand function on H(b)
for the ith commodity in some two person economy. In the Appendix we establish
a representation of polynomials in several variables as the sum of powers of polynomials that are linear in the variables. This combines with Lemma 1 to yield
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
552 HUGO SONNENSCHEIN
Theorem 2, which is the representation of an arbitrary polynomial on H(b) as an
excess demand function on H(b). In Theorem 3 we analyze the relationship between
the excess demand functions for the ith and jth commodities implied by the
techniques of construction employed through Theorem 2. It is shown that the
method is capable of generating precisely those polynomial excess demand functions
which satisfy the symmetry condition that the change in excess demand for the ith
commodity (1 < i < n - 1) with respect to a change in thejth price (1 < j < n - 1)
is equal to the change in excess demand for the jth commodity with respect to a
change in the ith price (the changes in excess demand for the nth commodity are
determined by Walras' Law). This result enables us to prove Theorem 4: Let
p1 p2,... , pm be a finite set of points in H(b), 6 > 0, that have no co-ordinates
in common. There exists an economy which has an equilibrium at each pJ, j = 1,
2,.. ., m, and none elsewhere in H(b). Finally, by alternating our choice of numeraire, and appealing after each alternation to a very weak form of Theorem 3, we
obtain Theorem 5: Walras' Law is the only restriction on the relationship among
a given price vector pi, excess demand at i, and the rates of change in excess demand
with respect to price changes evaluated at p.
4. THE TWO COMMODITY CASE
THEOREM 1: If f has a continuous derivative on [6, 1/6] for some 6, 0 < 6 < 1,
then there exist two consumers (U', wl) and (U2_ )2) such that f = EJ on [6, 1/6],
J = {1,2}.
PROOF: Write f = 4 + / where q(pl) = ap, - b, a > 0, b > 0, 2a/b < 6, and
/ is positive, strictly decreasing, and has a continuous derivative on [6, 1/6].
We now will construct two individuals whose excess demand functions on [6, 1/6]
are / and 4 respectively. This is all that is necessary, since the excess demand
function generated by the union of two individuals is the sum of their excess
demand functions.
We turn our attention first to /. Let wI be any positive number, and
09 - pj/(pj) > I for all p, E [6, 1/6]. This is done to guarantee that the demand
for the second commodity by the first person is positive. Using the fact that /
satisfies a uniform Lipschitz condition on [6, 1/6], one can prove that there exists
an e > 0 and an ij > 0 defined on [6, 1/6] such that no point of the form
(w)I + Ip4(P), - )2 p )) (h,(p,),h2(p1)) lies in S(Pj) = {(x1, x2):(P, - i7)x1 +
X2 < (P1 - t)h (j1) + h2(Pi)} whenever 0 < (Pr - -1) < E. Now define Ii =
min {6, tj}/2. The utility function U' is obtained in the following manner:
Given a point (hl(pl), h2(fil)) on a = {(hj(pl), h2(pl)): Pl E [6, 1/6]}, let the points
defined in (7a), (7b), and (7c) be assigned the value hl(pl) under U':
(7a) A(pl) = {(hj(pl), h2(p,))}
(7b) B(Pj1) = {(x1,x2):x1(p1 - j) + x2 = h,(p )(j,-j) + h2(01),
0 < X2 < h2(i)},
(7c) C(hi,) = {(x,,x2):xl = h,(ji)andx2 > h2(,)}.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
DEMAND
]FUNCTIONS
553
(The need for ji > 0 in (7b) is to guarantee that the portion of the indifference
curve below h2(fl) can be chosen so that (hl(pl), h2(jl)) is the unique choice at
prices jil.)
It must be shown that U' is well defined where it has been defined. This will be
established if the segments of the iso-U' contours below the path a do not cross,
and this will be the case if a r- B(jil) is empty for each j1 E [6, 1/6]. The latter fact
follows from the mean value theorem and our choice of ,t.
U' is continuous, satisfies (5) and (6), and is defined on a subset of the commodity
space which is bounded above by the iso-U1 contour corresponding to U1 = h1(b)
and below by the iso-U1 contour corresponding to U' = hl(l/b). It can readily
be extended continuously to the entire northeast quadrant in such a way so as to
preserve the satisfaction of (5) and (6).
As a consequence of this construction, for all j1 e [6, 1/6], (hl(pl), h2(4l)) is the
unique U' maximizing bundle corresponding to j1. We have thus found a consumer (U', w') such that / = El on [6, 1/6]. Next we must construct a consumer
(U2, w2) such that =E on [6, 1/6]. Let c > sup {+(x) :6 , x < 1/6}, )2 = 1,
and note that g(pl) =-pl (pl) is an increasing function of p, on [6, 1/6]. Let
P2 = l/pl. Observe g is a positive decreasing function of P2 and has a continuous
derivative. By the method employed to construct the first individual, there exists
a second individual such that g(p2) = E2(p2) on [6, 1/6]. Thus (- 1/pl)(1/pl) =
g(1/pO) = g(p2) = E2(p2) = E2(1/p ) = (- 1/pl)El(1/pl), and so 4(pl) = El(pl) on
[6, 1/6]. Since f = / + 4, this completes the proof of the theorem.
X2
W2
P)
h2\ 1)
,x,, hiPIG 1)
FIGURE 1.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
554
HUGO
SONNENSCHEIN
The following two corollaries characterize, for the case of two commodities,
the structure of excess demand functions on compact subsets of price space. The
first corollary is exact, and a proof is obtained by using the fact that the excess
demand functions generated in Theorem 1 satisfy Walras' Law. The second
corollary states that an arbitrary continuous function and its companion, which
is determined by Walras' Law, are approximately excess demand functions for
the first and second commodities in a two commodity economy. This result is an
immediate consequence of Theorem 1 and the Weierstrass Approximation Theorem.
COROLLARY 1: If El has a continuous derivative on [6, 1/6] for some 0 < 6 < 1,
(El, E2) = E: [6, 1/6] -+ R2, and (p,, 1) - E(pl) = 0 for all P, E [6, 1/6], then there
exist two consumers (U', w') and (U2, w2) such that E = EJ on [6, 1/6], J = {1, 2}.
COROLLARY 2: If (El, E2) = E: [6, 1/6] R2 is continuous for some 0 < 6 < 1
and (p, 1). E(p1) = O for all P1 E [6, 1/6], then there exists two consumers (U', w 0')
and (U2, w2) such that IEj(p,) - E/(p1)j < 6 for all P1 E [6, 1/6] and i E {1, 2} = J.
5. EXCESS DEMAND FOR ONE COMMODITY AS A FUNCTION OF
MORE THAN ONE RELATIVE PRICE
Lemma 1 sets the stage for the main result. It proves that any function g of
n - 1 prices, which can be represented as a continuously differentiable function f
of a linear function L of the n - 1 prices on H(6), is the excess demand function
on H(6) for the jth commodity in some two person economy. The technique of
proof is to define a composite commodity, whose unit cost at p E H(6) is L(p), and
to apply Theorem 1 to construct a two person economy whose demand for the
composite commodity is f (L(p)).
LEMMA 1: Let g be defined on H(6) and 0 < 6 < 1. If there exist n - 2 numbers
, 2.a2, ,- 1 aj+ *. ,cOtne- between zero and one, and a continuously differ-
entiable function f such that g(PI,P2, . . Pn-l) = f (?xP, + ... + ?cj-lPj-I +
pi + Xi+ lPj+ 1 + * * + (Xn- lPn- 1) for all (P1P2, . . , Pn - 1) E H(6), then there exist
two individuals (U', w)') and (U2, w2) such that Ej(pl, P2, , Pn- 1) = g(Pl, P2,
Pn- 1) for all (Pl, P2, .. ., Pn- 1) E H(6), where J = {1, 2}.
PROOF: Define a composite commodity xc to be one unit of commodity j and
ak units of commodity k, k = 1,2, ... j- 1,] + 1 .. . ,n - 1. Let p, = alp, +
* + cxj-1Pj- + pjti a+ Pi+1 + *.. + ,n-lPn -; Pc is the price of one unit of
commodity xc. By Theorem 1 there exist two consumers (U", wt)") and (U2', w)2')
such that E"'(pc) = f (pc) for all PC E [6/(n - 1), (n - 1)/6] = [6', 1/6'], where J' =
{ 1', 2'). Ec' may be viewed as taking values in S' = S nr Q, where S is the subspace
of Rnwhich is spanned by(cx,x2,. . . c a2 , xi, - 1 . . . I 1 0) and (, ,, . ., 1).
The utility functions UX and UT' are defined only on S'. In order to establish the
lemma they must be extended to functions U' and U2 defined on Q in such a
manner so that Ej(Pl,P2J*--Pn-l)=g(Pl,P2 Pn-1) for all (P1,P2,--.,
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
DEMAND FUNCTIONS 555
Pn- 1) E H(b), where J = {1, 2}. This is achieved in the following manner. For any
point x = (x1, x2, . . ., xn) E Q, let m = min {Xll X2/a2, - * xj, * Xn- 1/n- 1}
and let the Ui utility of x be the Uc utility of +(x) = (c^2m, 2m,... ., cXj 1m, m,
cXj+ 1m, ... , acn 1m, xM), which is defined since +(x) is a point in S'. Finally, let
09i = (ow1 w)Ci a2044. c cjJ l , c l aj+ 1wci, ..., I c4 wl' a). Consider now (Pl, P2,
I .,pn-D)eEH(b), and note that (P1,P2, .Pn -p1)EH(O) implies (cxlP1 + +
aj- Pj- 1 + Pj- 1 + aj+ 'Pj+ 1 + . . . + gn- 1Pn- 1) E [6' 1/6']. Individual i will maximize utility over the set {(X1, x2,.. X , Xn) :plXl + P2X2 + + Pn- lXn- 1 + Xn <
P1JW) + P2w + *.. + Pn1w 1 + wc} i = 1, 2. But as a consequence of our con-
struction this is equivalent to maximizing U' on {(xc, xn) :pcx + xn i pca4 + w9 },
EJ(Pj1 P2 ** Pn- 1) = Ec(pl E P2 ** Pn- 1)
i= 1,2. Thus, since each unit of the composite commodity c contains one unit of
commodity j,
= Ej (1p + ... + ij-lPj-1 + Pj
+ aj+ 1Pj+ 1 + + an-lPn-1)
= f (PC)
= g(Pl, P2,* Pn. )
for all (Pl, P2,** * , Pn- 1)7J(46)
The next result is fundamental for our analysis. Given an arbitrary polynomial
g defined on H(b), 0 < 6 < 1, and a number j, 1 < j < n - 1, we will show that
there exists an economy such that g is the excess demand for the jth commodity
on 17(b).
THEOREM 2: If g is a polynomial defined on H(6), and 0 < 6 < 1, then there exists
a collection of individuals {(Ui, o_)')j, i E J, such that EJ = g on H(6).
PROOF: Assume that g is of degree q > 0, and let 0 < cxo < cx1 < c2 <... <
osq < 1. By the theorem proved in the Appendix, g may be written as the sum of
X (i + 1)n2 terms of the form
(8) c(dlpl + d2p2 + . + d_lPj- 1 + Pj + * + dn-lPn-)i,
where the above sum is taken over 0 < i < q, and the dk's are chosen from
{ Ia, I,... ., Xq}. By Lemma 1 each term of this form can be generated as the
excess demand function for the jth commodity on H(b) by two individuals. Since
the excess demand function generated by the union of economies is the sum of
their excess demand functions, there exists an economy with 2 z (i + )n- 2
individuals which has g as excess demand function for the jth commodity on H(6).
(In forming the union of economies, individuals with identical tastes and initial
holdings are treated as distinct points.)
By applying a theorem on simultaneous interpolation and approximation [11],
one obtains the following corollary:
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
556 HUGO SONNENSCHEIN
COROLLARY 3: If g is an arbitrary continuous function defined on H, 0 < 6 < 1,
and p(l), p(2),. . . , p(m) are m points in H(b), then there exists a collection of indivi-
duals {(Ui, o_)'), i E J, such that g(p(k)) = EV(p(k)), k = 1, 2, ... , m, and lg(p) - Ejj(p)l < 6
for all p E H(b).
6. EXCESS DEMAND FOR SEVERAL COMMODITIES
So far we have been concerned with the following problem: given an arbitrary
function of n - 1 variables, is it an excess demand function? We now ask a more
difficult question: given n - 1 functions, each of n - 1 variables, are they excess
demand functions for the first n - 1 commodities in an n commodity economy
(the excess demand for the nth commodity is of course determined by Walras'
Law)? In Theorem 3 we provide an affirmative answer for a rather restricted class
of functions.
THEOREM 3: Let 0 < 6 < 1, and E1l E2, . .. En1 be n - 1 polynomials defined
on H(6) satisfying aE,/ap, = aE,/ap, (s, t = 1, 2, ... n - 1). Then there exists an
n commodity economy {(Ui, wi)}, i E J, such that (EJ, EJ2, ... , Ej- ) = (E1, E2,....
En- 1)on H(6).
PROOF: In order to simplify the exposition we will prove this theorem for n = 5;
the argument is easily generalized. We also note that it is sufficient to prove this
theorem for the case of the given polynomials homogeneous of degree m. Thus we
write
Es(P1 E P2 E P3 E P4) = E ppp, i +j + k <m,
i, j k >, O, s = 1, 25 35 45
and observe by the symmetry condition that
(9.1)
Wi1jk
=
ak
-
a(i
-
1)jk
(m - i -j- k,i,j,k (m-(i- 1)-j-k,i- - 1,j,k
= 2
= (i- 1)jk
(9.2) w[, __j_i a(j- 1)k
m - i-j- k,i,j,k m - i -(j - 1)- k,i,j - 1,k)
W3
wi(j1)k 5
(9.3)
Wi 1~ajk
a4-1 )
ij(k-
m - i-j- k,i,j,k m - i i - (k7- 1),i,j,k -I
=wij(k-1) 1
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
DEMAND
FUNCTIONS
557
94))jk2
3 )k,
(9.4) W(i= W(jl
(9.5) w = Wi(k-1), and
(9.6) wlil)k = W 1)
where the symbols Wjk are defined in (9.1>(9.3).'
Let %'0, 71,.. ,?m+ 1 be m + 2 distinct numbers all of which are greater than
zero and less than one. Consider the system
a mi- -k i i k
i+j+k(<m
- Z Z Cuqr(Pl + (XuP2 + OCqP3 + (rP4) ,
u=O q=O r=O
a 2 m-i-j-k ij k
Z iakpi1 P2P3P4
i+j+k(m
(10.2) i,j,k + + 0
m+l m+lm+l
- Z Z auCuqr(Pl + OuP2 + CLqP3 + CXrP4)m,
u=O q=O r=O
z aijkppm-i-jP2p3p4
(13) i j,k >, O
m+l m+lm+l
Z Z Z aqcuqr(Pl + auP2 + (XqP3 + (XrP4) ,
u=O q=O r=O
E 4 m-i- J-k i i k
z ajkP1lt'P2P3P4
i+j+k-<m
(10.4) i,j,k) o
m+ 1 m+ 1 m+ 1
- E E E crCuqr(Pl + (XuP2 + cLqP3 + (rP4)in
u=O q=O r=O
Expanding the above expressions by the multinomial formula yields the fact that
(10) will be satisfied if there exists {Cuqr}, 0 < u, q, r ? m + 1, such that
m+1 m+1 m+1
tjk
u q r= 'Iik
(11.1) W,.
= cuqr
E i urjoek
- tzjk,
u=0 q=0 r=0
m+ 1 m+ 1 m+ 1
(11.2) Wik = Z Z Z Cuqr Ilu = 1 l2jkk
u=0 q=0 r=0
m+1 m+1 m+1
(11.3) W,.tjk
= Z ,
'5' uqr
'' c L uq
iqi+lak
= 03k
r tk
u=0 q=0 r=0
m+1 m+1 m+1
(11.4) w 4j = c (x , cc c"ruiak + = 0j,~ ijk >, O,i +j + k < m + 1
u=0 q=0 r=0
4Assume that (9.1H9.3) are satisfied and i + j + k = m + 1. The argument W(i _)jk = W-k =
W3u_l)k is invalid since wb.,k is not defined. Thus (9.4) is not implied by (9.1H9.3). Likewise (9.5) and (9.6)
are not implied in such a manner.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
558 HUGO SONNENSCHEIN
Let {WHk} (s = 1, 2, 3, 4) be defined by the right hand equalities and observe that
(12) q50jk = (i-l)jk =4 -1)k = ij(k-1) for all i,, .
Define whk for i + j + k =m + 1, i,j, k O, by
(13) Whjk = W(i-1)jk'
and recall from (9.4H9.6) that W21)jk=W3()jk = W4j(k 1t follows from (9)
and (12) that a solution to (11.1) for all i,j,k > 0 satisfying i +j + k < m + 1
is also a solution to (11.2), (11.3), and (11.4). For if (11.1) is satisfied, then
/3
y
Wijk = W(i + 1)jk = (i + 1)jk = kijk,
(14) Wijk = Wi+ 1)k = iUj+ 1)k = Pijk
Wijk = Wj(k + 1) = lj(k+ 1) = kijk5
for all i,j, k > 0, i + j + k < m. The equalities below a follow from (9) and (13),
the equalities below ,B follow from (11), and the equalities below y follow from (12).
However, (11.1) has a solution (see Lemma, Appendix), and thus there exist
(m + 2)3 real numbers {Cuqr} which make all of the equations in (11) hold.
Consider now the (m + 1)3 terms (each of the form cuqr(Pi + auP2 + aqP3 +
OrP4)m) in the sum which forms the right hand side of (10.1). For each of those
terms there exists, by Lemma 1, two consumers who generate that term as excess
demand for the first commodity on HI(5), n = 5. An investigation of the proof of
Lemma 1 reveals that these consumers generate
aucuqr(Pl + auP2 + aqP3 + arP4)m, aqCuqr(Pl + CUP2 + aqP3 + OcrP4)m,
and
arcuqr(Pl + auP2 + aqP3 + arp4)m,
as excess demand for the second, third, and fourth commodities respectively.5
Thus, the economy {(Ui, Wi)}, i E J, composed of two individuals of the type constructed in Lemma 1 for each of the (m + 2)3 terms in the sum in (10.1), has the
functions on the right hand side of (10) as excess demand functions for commodities
one, two, three, and four. But we have shown that (11) is satisfied, which implies
that the equations in (10) are satisfied, and yields
E(pl, P2, P3, P4)= EE Cuqr(Pl + auP2 + (XqP3 + arP4)m
u q r
a= ,- itkP j2P k= Ej(pj, P2 P3, P4),
EJ(pl, P2, P3, P4) Y l aucuqr(Pl + auP2 + aqP3 + cXrP4)m
u q r
=Z aukpi p2pJ3p4 = E2(P1 P2, P3E P4),
For each unit of commodity one, they demand a,, a., and a, units of commodities two, three,
and four.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
DEMAND
FUNCTIONS
559
E3(pl, P2, P3, P4) = E aqCuqr(Pl + auP2 + aqP3 + arP4)m
u q r
= Z ajkpl P2P3P4 = P1, P2, P3 P),
E4(pl, P2, P3, P4) = Z Z Z arCuqr(Pl + auP2 + aqP3 + arP4)m
u q r
= Z a 'kPi P2P3P4 = E4(pl, P2, P3, P4)
on H(3) (n = 5). This completes the proof of Theorem 3.
It is a well known fact that not all excess demand functions can be linear in
prices. An immediate consequence of the preceding analysis is that n - 1 of them
can in fact be linear and non-constant in the n-I prices.6
This is apparent from the following argument: Consider the quadratic form
n-1 n-1
F(p1,p2P, Pnp - )= Z aijipZ
and the functions aF/ap, a ,F/Ip2 ... ,a F/Pn- 1. These n - 1 functions are linear
and satisfy the conditions of Theorem 3 (since 02F/@piapj = 02F/Opjapi). Thus
there exists an economy which generates them as excess demand functions for the
first n - 1 commodities.
Unfortunately our methods will not work to construct an economy which
obtains arbitrary functions of the form
E(P1, p2) = alPl + a2P2 + a3'
E2(pl, P2) = f,lPl + f2P2 + /33,
as excess demand functions for the first two commodities in a three commodity
economy. The fact that it is unknown whether or not there exists an economy
which yields arbitrary excess demand functions of this form provides a most
powerful illustration of our lack of knowledge concerning the structure of excess
demand. We now turn to an investigation of what "equilibrium price sets" can
look like.
7. EQUILIBRIUM PRICE SETS
For a given economy {(Ui, coi)}, i E J, value is determined by those price vectors
in a (or H) which are mapped into the zero vector by the function EJ. It is well
known that in general more than one price vector will be mapped into the zero
vector; i.e., value will in general not be uniquely determined by a static specification
of an economy. Theorem 4 determines a class of subsets of H(b) which are possible
equilibrium price sets for an economy.
6 The question of whether or not this is possible was brought to my attention several years ago by
James Quirk and Rubin Saposnik.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
560 HUGO SONNENSCHEIN
THEOREM 4: Let p', p2, ... , pM be M arbitrary points in 1(b), 3 > 0, which have
no co-ordinates in common. There exists an economy which has an equilibrium at
each pi, j = 1, 2, .. ., M, and none elsewhere in H(b).
PROOF: Consider the polynomial
j=M i=n-1
F1(p) = H Z (Pij=1 i=1
This function is non-negative and is zero if and only if p E {pi}. By Theorem 2 there
exists an economy {(Uj, coi)}, j E J(1), which has F1 as excess demand for the first
commodity on 17(b).
Consider the M numbers {-Ej(1)(pj)}, 1 j <I M, and let F2 be a polynomial
in P2 only which takes on the value - Ej(1)(pi) at pi, 1 < j < M. This is possible
since the pJ have been chosen so that pk = pij,j =? k. By Theorem 3 there exists
an economy {(Ui, coj)}, j E J(2) such thatEJ(2) = 0 on H(b) if s = 2, n, and EJ(2) = F2
on H(b). Let F3 be a polynomial in P3 only which takes on the value3-EJ(1(p1) at
pi, 1 < j < M. This is possible since the pi have been chosen so that pk =: pj 'j = k.
By Theorem 3 there exists an economy {(Ui, wo')}, j E J(3) such that EJ(3) = 0 on
H(b) if s = 3, n, and EJ(3) = F3 on H(b).
Continue this process. The economy {(Ui, co)}, j e u J(i) where the union is
taken over 0 K i < n - 1, has an equilibrium at each point pi, 1 < i < M; it can
have no other equilibria in H(b), since the excess demand for the first commodity
is not zero anywhere else in H(b). This completes the proof. The reader will note
that an analog of this result obtains if I7(b) is replaced by A(b).
8. A LOCAL RESULT WITH AN APPLICATION TO STABILITY ANALYSIS
The next theorem establishes that at any given price vector, the first n - 1
(of n) excess demand functions and their partial derivatives may take on any
arbitrary set of values. This is informally summarized by saying that Walras' Law
is the only local comparative statics theorem of competitive equilibrium analysis.
THEOREM 5: Assume p = (P1, P2, ... ., Pn-1 ) belongs to H, (a(j, k)) is an arbitrary
(n - 1) x (n - 1) matrix, and N(1), N(2), .. ., N(n - 1) are arbitrary numbers, then
there exists an economy {(Ui, coi)j, i E J, such that EJ(p) = N(j) and aEj(p)/aPk =
a(j, k),1 < j, k < n -1.7
PROOF: The proof is given only for n = 3 since this illustrates the general
method. If we consider commodity two to be the numeraire, i.e., if we fix its price
at one, then by Theorem 3, there exists an economy (composed of individuals
indexed by J(2)) which generates c1, -c1p1 - C3p3, C3 as excess demand for
7 It was demonstrated to me by Professor Paul Samuelson, that for a single individual, the matrix
(aEj/aPk - aEJkpj) is the sum of two matrices of rank one, and hence (aEj/ap) is not arbitrary. This
provides the basis for a rigorous distinction between the single consumer and many consumer cases.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
DEMAND FUNCTIONS 561
commodities one, two, and three respectively. If P2 is entered as a variable, then
the excess demand function generated by {(Ui, wi)}, i E J(2) must be homogeneous
of degree zero in Pl, P2, and p3, and agree with the non-homogeneous form when
P2 = 1. It is thus c1, (- c1pl - c3p3)/p2, c3. Likewise, there exists an economy
(composed of individuals indexed by J(3)) which generates excess demand
(- d2p2 - d3p3)/p1, d2, d3, and an economy (composed of individuals indexed
by J(1)) which generates excess demand el, e2, (- elp, - e2p2)/p3. The excess
demand for the first two commodities generated by {(Ui, wi)j, i E u J(k) is
e1 + c1 - (d2p2 + d3p3)/p1, e2 + d2 - (c1pl + c3p3)/p2. Setting P3 = 1 yields
(15) e1 + c1 - (d2p2 + d3)/p1, and
(16) e2 + d2 - (c1pl + c3)/p2,
as excess demand for commodities one and two at prices Pi and P2. The result
follows from first differentiating the above expressions partially with respect to
Pi and P2, and then noting that by appropriately choosing the c's, d's, and e's,
both (15) and (16), and the four expressions derived by differentiation, may be
assigned any desired value at any point in H, n = 3.
One use of Theorem 5 is to help answer questions concerning the relationship
between various definitions of local stability. For example, although it is known
that there are matrices that are Hicksian stable but not local dynamically stable,
it has not been shown that there are economies which yield these arrays as matrices
of aEJ/apj's at an equilibrium [5, pp. 188-9, and 8]. This problem is resolved in
Corollary 4.
COROLLARY 4: There exists an equilibrium that is Hicksian stable, but not
dynamically stable.
PROOF: The matrix
-1 +1 +14
-1 -1 +1
_ O -1 -1_
is Hicksian stable, but not dynamically stable [5, p. 173]. By the previous theorem
there exists a four commodity economy which has an equilibrium at prices (1, 2, 3).
and which generates excess demand functions whose jth co-ordinate function
differentiated with respect to Pk, and evaluated at (1, 2, 3), is precisely the j, kth
entry of the above matrix, 1 < j, k < 3. Such an economy has an equilibrium which
is Hicksian stable but not dynamically stable at prices (1, 2, 3).
9. UNSOLVED PROBLEMS
All of the preceding theorems would be implied by an affirmative answer to
Question A: Can any n - 1 functions in n - 1 prices, which have continuous
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
562
HUGO
SONNENSCHEIN
derivatives on H(b), be realized as excess demand functions for the first n - 1
commodities in an n commodity economy?8
Also interesting is Question B: If Question A is answered in the affirmative,
what additional restrictions must be imposed on the given functions so that when
H(b) is replaced by H (or A) in A, an affirmative answer still obtains?
University of Massachusetts at Amherst
Manuscript received February, 1970; revision received December, 1970.
APPENDIX
A REPRESENTATION FOR POLYNOMIALS OF SEVERAL VARIABLES
The purpose of this appendix is to prove the following:
THEOREM: Any polynomial in several variables can be written as a sum of powers of linear polynomials.
More explicitly, let integers m > 1, n > 0, and distinct numbers a(0), a(1), . . ., a(n) be given. Consider
the K (n + 1)' linear polynomials Z1, Z2, . ., ZK of the form xo + ci(s)x1 + 4(t)x2 + . . . + c(u)xn,
0 < s, t. . , u < n. Then the collection {(Zj)n: j = 1, 2,..., K} spans the vector space of homogeneous
polynomials of degree n in the (m + 1)-variables xo, x . . ., Xm
The proof of the Theorem is based on the following:
LEMMA: If a(0), a(1), . 4. , a(n) are n + 1 distinct real numbers, and {w(i, j, . . k): 0 < i, j, . . k < n},
are real numbers, then there exist real numbers {c(s, t, . . ., u): 0 < s, t, . . ., u < n} such that w(i, j, . , k) =
y c(s, t . . , u)a(s)io(ty ... OC(u)k for all 0 < i, j, . . ., k < n, where the above sum is taken over {s, t, . u:
0 < s,t,...,u < n}.
PROOF: Let m denote the number of arguments of w. For m = 1 the result follows immediately
from the non-singularity of Vandermonde matrices. We will prove the lemma only for m = 2, since
the procedure employed is illustrative of the proof of the necessary induction step.
If a(0), a(1), . . ., a(n) are distinct real numbers, and {w(i, j): 0 < i, j < n} are real numbers, then we
must find real numbers {c(s, t): 0 < s, t < n} such that
n
n
n
n
w(i,j) = si ?9 c(s, t)c*(s)i0c(ty = s 0 c(s, t)ae(ty 0(s)i
s=O
t=O
s=O
t=O
for all 0 < i,j < n.
Since the lemma holds for m = 1, for each j, 0 < j < n, there exist n + 1 real numbers du, k),
0 < k < n, such that w(i,j) = Y d(s,j)c*(s)' for all 0 < i < n, where the sum is taken over 0 < s < n.
The proof (for m = 2) is completed by applying the lemma (for m = 1) n + 1 times to obtain real numbers
c(s, t) which satisfy d(s, j) = Y c(s, t)ce(t)j for all 0 < s, j < n, where the sum is taken over 0 < t < n.
PROOF OF THE THEOREM: We will prove that the collection {(Zj)n :j = 1, 2, . . ., K} spans the vector
space of homogeneous polynomials of degree n in three variables x, y, and z, since this illustrates the
general case of m + 1 variables. If P is homogeneous of degree n in x, y, and z, then
(Al) P(x, y, z) = E a(i, j)x n-i- jyizi,
where the sum is taken over S = {(i,j):0 < i,j < n, i + j < n}. For 0 < i,j < n, let
a(i,j) ifi+
w(i j) = |
n n-l- J, l,J
tn if i +j> n.
8 Added in proof: The author obtained an affirmative answer to Question A for the case of polynomial excess demand in January, 1972.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms
DEMAND
FUNCTIONS
563
We must show that
(A2) P(x, y, z) = Z c(s, t)(x + a(s)y + at)z)"
= Z C(S, t) n _ ) _ i (S)i(tyXn-i- jyizi)
for some collection of real numbers {c(s, t) :0 < s, t < n}, where the sum inside the brackets is taken
over S, and the sum outside is taken over T = {(s, t):O < s, t < n}. Matching coefficients in (Al) and
(A2) we conclude that the Theorem will be established if we can find {c(s, t):0 < s, t < n} such that
w(i,j) = Y c(s, t)4(s)'#t)j, for all 0 < i,j < n, where the sum is taken over T But this follows from the
lemma with m = 2.
An alternative proof can be found in [10].
REFERENCES
[1] DEBREU, G.: "Economies with a Finite Set of Equilibrium," Econometrica, 38 (1970), 387-392.
[2] GALE, D.: "The Law of Supply and Demand," Mathematica Scandinavica, 3 (1955), 155-169.
[3] HURWICZ, L., AND H. UZAWA: "On the Integrability of Demand Functions," in Preferences,
Utility, and Demand, ed. John Chipman, et al. New York: Harcourt Brace Jovanovich, Inc.,
1971, 114-148.
[4] NEGISHI, T.: "The Stability of Competitive Equilibrium," Econometrica, 30 (1962), 635-670.
[5] QUIRK, J., AND R. SAPOSNIK: Introduction to General Equilibrium Theory and Welfare Economics.
New York: McGraw-Hill, 1968.
[6] SAMUELSON, P.: The Foundations of Economic Analysis, Cambridge: Harvard University Press,
1947.
[7] : "The Problem of Integrability in Utility Theory," Economica, N.S., 17 (1950), 355-385.
[8] "The Relation Between Hicksian Stability and True Dynamic Stability," Econometrica,
12 (1944), 256-257.
[9] SCARF, H.: "Some Examples of Global Instability of Competitive Equilibrium," International
Economic Review, 1 (1960), 157-172.
[10] SONNENSCHEIN, H.: "A Representation for Polynomials of Several Variables," American
Mathematical Monthly, 19 (1970), 45-47.
[11] YAMABE, H.: "On an Extension of the Helly's Theorem," Osaka Mathematics Journal, 2 (1950),
15-17.
This content downloaded from 81.194.22.198 on Sun, 03 Jul 2016 17:28:49 UTC
All use subject to http://about.jstor.org/terms